Add HUMANML3D_VIDEOMDM_ON_MVLIFT

HUMANML3D_VIDEOMDM_ON_MVLIFT/args.json

@@ -0,0 +1,80 @@
+{
+    "adam_beta2": 0.999,
+    "arch": "trans_enc",
+    "avg_model_beta": 0.9999,
+    "batch_size": 64,
+    "cam_sample_distance_factor": 2.5,
+    "cfg_for_distilation": 0.0,
+    "clip_grad_max_norm": 1.0,
+    "cond_mask_prob": 0.1,
+    "cuda": true,
+    "data_dir": "",
+    "dataset": "video_mdm_synthetic_mvlift",
+    "device": 0,
+    "diffusion_steps": 50,
+    "distilation_branched_denoising": false,
+    "distilation_solve_ode": false,
+    "distilation_use_gt_camera": false,
+    "dont_use_ema": false,
+    "emb_trans_dec": false,
+    "eval_batch_size": 32,
+    "eval_during_training": true,
+    "eval_num_samples": 1000,
+    "eval_rep_times": 3,
+    "eval_split": "val",
+    "evaluator_dir_path": "",
+    "external_mode": false,
+    "fine_tunning": true,
+    "gen_during_training": false,
+    "gen_guidance_param": 2.5,
+    "gt_supervision": false,
+    "gt_teacher": false,
+    "gt_xyz_supervision": false,
+    "lambda_cam": 1.0,
+    "lambda_cam_complement": 0.0,
+    "lambda_cam_vel": 258.0,
+    "lambda_consistancy": 1.0,
+    "lambda_fc": 0.0,
+    "lambda_floor_distance": 0.0,
+    "lambda_rcxyz": 0.0,
+    "lambda_sds": 0.0,
+    "lambda_sds_vel": 50.0,
+    "lambda_vel": 0.0,
+    "latent_dim": 512,
+    "layers": 8,
+    "log_interval": 1000,
+    "lr": 0.0001,
+    "lr_anneal_steps": 0,
+    "mask_frames": false,
+    "mask_from_score_type": "square",
+    "max_cam_sample_elevation_angle": 0.1308996938995747,
+    "min_cam_sample_elevation_angle": -0.39269908169872414,
+    "model_2d_path": null,
+    "multistep_use_guidance": false,
+    "no_distance_weighting": false,
+    "noise_schedule": "cosine",
+    "num_cameras_for_distilation": 1,
+    "num_frames": 60,
+    "num_steps": 200000,
+    "num_workers": 2,
+    "overwrite": false,
+    "pnp_to_find_cameras": false,
+    "pos_embed_max_len": 5000,
+    "quicker_batch_chance": 0.0,
+    "resume_checkpoint": "",
+    "sanity_check_folder": null,
+    "save_dir": "save/HUMANML3D_VIDEOMDM_ON_MVLIFT",
+    "save_interval": 50000,
+    "seed": 10,
+    "sigma_small": true,
+    "t_star": 30,
+    "t_star_arg": 3,
+    "t_star_method": "multistep_b1_e1_d1",
+    "text_encoder_type": "clip",
+    "train_camera_azimuth_corruption": 0,
+    "train_platform_type": "WandBPlatform",
+    "train_split": "train",
+    "unconstrained": false,
+    "video_mdm": true,
+    "weight_decay": 0.0
+}

HUMANML3D_VIDEOMDM_ON_MVLIFT/model000600091.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9996644fcba21510f68bca167d8e008e7ecba99eaa2b72f51d7ea58bb72da060
+size 163642037